Roblox’s AI assistant gets new agentic tools to plan, build, and test games

Roblox has announced exciting updates to its AI-driven development tool, Roblox Assistant, which are set to revolutionize how creators design, build, and test games on its platform. These new features, aimed at enriching the development experience, were exclusively shared with TechCrunch. The revamped Assistant introduces a sophisticated 'Planning Mode' designed to serve as a collaborative partner for developers throughout the game creation process. According to Roblox, traditional AI tools that provide a solution based on a single prompt often miss the mark on capturing the creator's true vision. Planning Mode addresses this by analyzing game code and data models, asking insightful questions, and transforming prompts into actionable, editable plans. For instance, if a developer requests the creation of a park mini-game featuring a fountain and greenery, the Assistant might inquire about the desired visual style—options could include cartoony, realistic, or fantasy. Additionally, it may ask how to construct the park’s components, offering choices between building assets from scratch, sourcing them from the Creator Store, or a combination of both. After establishing a comprehensive plan, Planning Mode will utilize Roblox's suite of AI tools, including two newly introduced features: Mesh Generation and Procedural Model Generation, both aimed at accelerating the development timeline. Mesh Generation simplifies the integration of detailed 3D objects into the game environment. Traditionally, developers relied on placeholder assets during early stages, but with this new tool, they can swiftly generate high-quality 3D models. For example, a creator can instruct the Assistant to produce a campfire, enhance it with realistic lighting, and set the scene during nighttime. Moreover, the forthcoming Procedural Models will enable developers to craft editable 3D models using code and the Assistant's guidance. This innovative feature allows creators to manipulate attributes such as the number of shelves in a bookcase or the height of a staircase dynamically, creating versatile building blocks that can be refined and reused across projects. Nick Tornow, Senior Vice President of Engineering at Roblox, emphasized that these agentic features in Roblox Studio bridge the gap between creative ideas and execution. He stated, "Creating with Planning Mode and our Procedural Generation tools introduces a powerful new methodology for transforming concepts into playable games. The Assistant acts as a collaborative partner, expediting the journey from idea to implementation." As Planning Mode operates according to the established plan, it employs playtesting tools to analyze output logs, capture screenshots, and utilize keyboard and mouse inputs to evaluate design and gameplay. The system can identify bugs and provide feedback to the Assistant, enabling automatic corrections. Roblox further elaborated that these advancements will enhance the Assistant's ability to use agentic loops for testing various game elements, offering suggested solutions, and integrating findings into future planning processes. This creates a self-improving system that evolves over time. In addition, Roblox is exploring the possibility of enabling multiple AI agents to collaborate on tasks simultaneously, manage complex workflows in the cloud, and undertake activities such as coding, testing, and developing more lifelike game characters. The company aims to ensure that creators can easily incorporate third-party tools like Claude, Cursor, and Codex within Roblox Studio.